Prescott's performance dictated by cash, not cache

Decoding Intel's intent behind the Prescott marketing message is an old-fashioned game

Over the years, the Pentium name has been associated with a bewildering array of extensions. Who now remembers the Pentium Overdrive? The Pentium Pro? The Pentium II PE? But if there's anything more confusing than a random name change driven by inscrutable marketing departments, it's no name change at all when something big is happening. Intel has just brought forth the biggest technological upgrade for five years, with the 90nm Prescott finally appearing to take the place of the 130nm Northwood Pentium. Here's how to tell them apart: pay attention.

The old Pentiums are known by names like "Intel Pentium 4 processor with HT Technology 3.40 GHz", but the new ones are distinguished by an added E -- thusly, "Intel Pentium 4 processor with HT Technology 3.40E GHz". Got that? On no account confuse the "Intel Pentium 4 processor with HT Technology 3.40E GHz" with "Intel Pentium 4 processor with HT Technology 3.40 GHz EE", though, because that's the old technology with a bigger cache -- the Extreme Edition -- which is faster than the new version. Intel has said that it isn't ruling out making an Extreme Edition with the new technology -- it's hard to see how it could avoid doing this, once it starts ramping the Prescott core to speeds where the Northwood can't go -- and then we may see "Intel Pentium 4 processor with HT Technology 4E GHz EE". Or perhaps not: even Intel has problems with this pointless proliferation, and I don't think I've seen many documents from the company where it gets it right throughout.

This is prima facie bonkers: no other industry makes major product steps with so much incomprehensible nonsense attached -- and if they do, as Coke tried when it replaced Coca-Cola with 'new' Coca-Cola, there is nothing but pain in store. But then, there is no other industry like the processor industry.

At heart, this confusing transition is driven by Intel's deep desire to ditch Northwood as soon as possible for cost reasons: nothing to do with the flow of electrons, everything to do with the flow of dollars. At the end of any process's life, the chips are as big and as complex as they're ever going to get -- and in chip fabrication, the size dictates the cost. The cost per wafer is relatively stable, and the number of defects per wafer is also reasonably fixed. Given that each defect will take out just one chip, no matter how big or small that chip is, then the more chips per wafer you have the greater the proportion of working silicon you get at the end of the process.

Alas, the marketing message "Buy our new chips, they've got bigger margins" is less than captivating. You could drop the price of course, but you badly want to start to recoup all that investment in inventing new and better fab plants. There's no point in going for better margins if you don't get better margins: you can cut the price a bit to keep the competition on their toes, but you're in this business to make money.

People only buy new chips when they do something better than the old ones, and the new Prescotts do everything worse than the old ones. So, the best solution is to make sure you can't buy the old ones -- but this requires some finessing of the marketing message. Which, to the untutored, can look a lot like a burst of smoke and mirrors after which the lady in the box has been replaced by an elephant.

But why not wait until you can make the new chips faster?

The standard excuse for new architectures going slower is that the new ideas have more headroom -- wait for a bit, and you'll find Prescott going places that Northwood could never reach. Yet you can't abandon your old chips -- and if you cut their prices to reflect their near-obsolescence, what was the point of bringing out new chips with better margins? So, sweep out the old, bring in the new, and ride out the wave of complaints that things ain't what they used to be. It worked when the early Pentium 4s replaced the late Pentium IIIs, despite the lack of performance increase: it'll work this time.

The trouble with making a change that is no change is that it leaves many questions unanswered. Will the new Pentium Ms work so much better in the near future -- especially where power consumption is concerned -- that the Prescott P4s will look poor alternatives? Will pressure from AMD's 64-bit enhanced chips mean that a later revision of the Prescott will include similar features -- and if so, why buy in now? There is no way that Intel's public roadmaps can reflect these considerations: this is a company that was vigorously denying the existence of the Extreme Edition Pentium -- a server chip wearing a spacesuit and toting a blaster ray -- in the face of direct questioning an hour before it was launched. We cannot expect too much help in decoding the runes, but the answer will lie not so much in the technology involved as the old journalistic adage: follow the money.